search

SeanChenxy / HandMesh

MIT License
339 stars 69 forks source link

Key matching issue for MobRecon #85

Closed jinhong-ni closed 1 year ago

jinhong-ni commented 1 year ago

Dear Authors,

Thanks for open-sourcing your great work. I just trained MobRecon on another dataset, and wish to test it on FreiHAND via the script you provided. However, it seems like the models defined in 'mobrecon/models/mobrecon_ds.py' and 'cmr/models/mobrecon_densestack.py' have unmatched keys, namely the followings: Missing key(s) in state_dict: "backbone.conv.0.weight". Unexpected key(s) in state_dict: "backbone.reduce.0.weight", "decoder3d.de_layer_conv.0.weight". Could you please let me know how you match the key for your provided pretrained weights, which I assumed is also trained following the model defined in 'mobrecon/models/mobrecon_ds.py' but loaded with 'cmr/models/mobrecon_densestack.py' during test time. Thanks in advance!

SeanChenxy commented 1 year ago

Hi, cmr floder is the old-version code. Please use mobrecon/models/mobrecon_ds.py for both training and inference.